- Reliability
- 41.5%
- Response time
- 1.91s
- Cost / min
- n/a
Gemini 3.8 Live is ahead on success rate, agent responses, data accuracy, calls where the agent went quiet and response time, and GPT Realtime 2.1 Mini on reliability and interruption.
Last updated Methodology by Dileep Chagam
For the wider field, see our pick of the best voice AI APIs for real-time audio.
Scenarios the agent passed on every one of its 3 runs.
Both providers, every model: Google vs OpenAI
Each model on its own: Gemini 3.8 Live and GPT Realtime 2.1 Mini
GPT Realtime 2.1 Mini is more reliable, by 8.5 points. Gemini 3.8 Live passed 34 of 82 scenarios on all three runs, and GPT Realtime 2.1 Mini passed 41. A scenario only counts when every run of it passed.
GPT Realtime 2.1 Mini is more reliable, by 8.5 points. Gemini 3.8 Live passed 34 of 82 scenarios on all three runs, and GPT Realtime 2.1 Mini passed 41. A scenario only counts when every run of it passed.
For a voice agent, GPT Realtime 2.1 Mini, since it gets more scenarios right on every run. Gemini 3.8 Live is ahead on success rate, agent responses, data accuracy, calls where the agent went quiet and response time, and GPT Realtime 2.1 Mini on reliability and interruption.
Gemini 3.8 Live, by 5.9 points. Gemini 3.8 Live saved everything exactly right on 70.5% of the calls that save data, and GPT Realtime 2.1 Mini on 64.5%. On Booking calls, 78.4% against 88.8%. On Medicare calls, 50.0% against 0.0%. A call counts only when nothing is missing, wrong, extra or saved twice.
Yes. Gemini 3.8 Live starts replying 0.17s sooner at the median. Gemini 3.8 Live takes 1.91s at the median and 2.23s at p90; GPT Realtime 2.1 Mini takes 2.08s and 2.75s. Both are measured on the call audio, from the caller finishing to the agent starting to reply.
GPT Realtime 2.1 Mini, by 0.01 on a 0 to 5 scale. Gemini 3.8 Live scores 4.98 and GPT Realtime 2.1 Mini 4.99 for how well the agent handled being talked over.
Gemini 3.8 Live, by 0.4 points. The agent went quiet for 10 seconds or more on 2.0% of Gemini 3.8 Live calls and 2.5% of GPT Realtime 2.1 Mini calls.
There is no verified per-minute rate for Gemini 3.8 Live, so cost cannot be compared yet.
Cekura ran both as the whole agent in the same open-source Pipecat pipeline, with the same prompt, tools and connection. A simulated caller worked through all 82 scenarios on live calls, three times each, 3,936 calls across the benchmark. A call passes only when the agent said the right things and saved the right data.
Cekura runs these simulated calls on your agent and scores every one.